Comment by pixelpoet
10 days ago
Are these models still relevant for people outside the US? I get the impression we're stuck on GPT 5.5 and Opus 4.8 pretty much permanently now, and relying on Chinese models in future.
10 days ago
Are these models still relevant for people outside the US? I get the impression we're stuck on GPT 5.5 and Opus 4.8 pretty much permanently now, and relying on Chinese models in future.
Dont worry, chinese models will distill frontier ones, quite fast.
The excuse they give is borderline childish. I get the thing about slow rollout, make sure partners get to fix the bugs, etc...
But bad actors are hard working motivated entities with tens of thousand of fake ids, and american citizens working for them, for pennies.
All while the ones like or you sit at a crossfire which is borderline useless.
I cant wait to see what Qwen did with the massive distillation they made out of Opus 4.8 and Fable aka Mythos aka pretty sure they jailbroke it.
Anybody who used codex or claude for the last few months is sitting on directories full of session logs that -- with the right motivation and effort -- can be massages and used for fine tuning or reinforcement training on any large model.
This is nothing a few felony indictments can't fix.
Pretty sure Chinese police will not cooperate with a US indictment
4 replies →
If the US really cracks down on frontier model access, you'll see them make Chinese open models illegal. You might say "oh well, let them try", but they will just put direct and secondary sanctions on every company whose systems have used Chinese models in some way. They just have to make an example out of a few international companies and no one will dare to use Chinese frontier models, at least commercially.
> but they will just put direct and secondary sanctions on every company whose systems have used Chinese models in some way
What about this: companies stop providing AI tokens to their employees entirely and instead, give a monthly budget for developer tools? They can even go as far as saying "if we realise that you use Chinese AI, you will get a warning and then be fired".
It's not like one can identify code coming from Chinese AI, right? As long as a company doesn't pay for those subscriptions, it may just be the employees writing the code all by themselves :-).
Not yet, but I could see them requiring some sort of signing in the future.
1 reply →
This is why we can't decouple ourselves from the US fast enough.
Nobody is going to care in Europe but a handful of big corps with huge us contracts.
And good luck proving it.
in the extremely unlikely event that they do this, what will happen is that Chinese models will become "rebranded" with a wink and a nod by the token routers (at the very least, the non-US ones). there is a zero percent chance that corporations will not work around it if the models are good and cheap.
Like Cursor, which is pretty much repackaged Kimi K2.5, and Musk paid $60b for (lol).
Theoretically that gives edge to all other companied around the world though, no?
I don't think a critical mass of them will oppose the US. The most likely equilibrium is Chinese models being shut out of any US-aligned markets (i.e. Europe at the very least, also East Asia, etc.). Probably India, Russia, Brazil etc. will resist such pressure, but they are protectionist and resilient to trade wars anyway, at the expense of their own welfare of course.
8 replies →
> but they will just put direct and secondary sanctions on every company whose systems have used Chinese models in some way
Yes, and the rest of the world would just nod worriedly and go along with it, at massive cost to their economies, rather than treating it like the protectionism it is and responding to it with crippling counter-sanctions.
> If the US really cracks down on frontier model access, you'll see them make Chinese open models illegal.
We don't give a fuck about US laws - respectfully, the rest of the planet.
We're already sick of your shit and this will only add to it. Just look at the Iran shit show. What a joke. Ooooooo wooooooo sanctionzzz scary. Sanctions only work if they're enforceable.
You think the US can tell the rest of the world "we're the only ones allowed to use frontier models" and that the rest of the world will just comply? There's just no way. Not even close US allies would go along with that.
I have no idea, but how is it easy to know whether somebody used these models? They can be hosted even locally.
And the chinese will just lie down and take this without doing anything?
You shouldn’t build a business that relies on any of these models. It’s a geopolitical and sovereignty risk now. Someone could just rug pull your entire stack.
Not only that, but using Opus 4.8 [1m] right now outside the US, and suddenly I only have a 500k context window. I really hope this is just a strange Claude Code bug, but I had access to a 1 Million window before, and it wouldn't entirely surprise me if context window length becomes another US export restriction.
The Anthropic page here seems to say that Max users should have access to the full 1 Million window for 4.8:
https://support.claude.com/en/articles/8606394-how-large-is-...
I was already setting up my infra to experiment with GLM 5.2 and its 1 Million token window before this happened. I think I'm glad I did.
EDIT: Found a solution, seems Claude Code 2.1.193 (or an earlier version I didn't notice) changed default settings, so that if you have Autocompact turned on it occurs at 50% of the context window. If you turn off Autocompact, the full 1 Million context window is restored. Another example of Claude Code quietly changing default settings sigh
You want to compact early though as sending the whole chat you will end up with a lot of tokens not in the cache which 1. Costs way more and 2. Will slow the request down as it has to process it all.
I do agree in cases where I'm using API and not the subscription, this would be very costly via API. Not sure why the tokens wouldn't be in the cache though? Seems everything should be cached as long as I'm within the 1 hour caching window? If I'm wrong about how token caching works, I'm eager to learn!
My other concern is, it isn't really a 1 Million context window if we can only use the first 500k, right? But now that I've found that I can re-enable it, I'm happy.
I've previously had sessions go to 700k tokens and still be okay, though it does start drifting at that 700k point. I'm regularly at 300k with no problem.
Hopefully. I hope this eventually nudges my employer to use Deepseek or other Chinese models.
That’s assuming China would not start controlling the access to their models.
Chinese companies can make a killing selling on prem AI systems to the rest of the world now.
Big boxes with Huawei GPUs and Chinese open models to run inside your company without network access.
They could, but I can imagine if US keeps on blocking the cutting edge models, China would never ship the cutting edge models and would still make a killing shipping models that are powerful enough for most of business cases
6 replies →
China has no reason to do that. The US is freely handing them the international market for AI.
US just needs their internationally usable model to be better than China's. If China catches up, US starts releasing more powerful models.
3 replies →
Are they though? I see this as a precautious method by US to maintain AI model superiority so the Chinese companies cannot distill from the US frontier models. Let's see how fast Chinese models would improve without access to latest US models and if they keep on releasing open models
5 replies →
Why do people always fearmonger China speculatively doing things that the US is actively doing?